We hope you enjoyed the December 5 webinar on the fundamentals of network virtualization! The webinar covered different approaches to the innovative process and the most important things to think about when picking a network virtualization vendor or setting up a proof-of-concept.
SDNCentral’s own Roy Chua and Nuage Networks Senior Director of Marketing Houman Modarres led the webinar and took questions from participants afterward. Read that Q&A below, watch the full presentation, or check out the teaser video and other resources including:
- A podcast of the webinar.
- A PDF transcript of the webinar.
- A downloadable PDF of presentation slides.
How is SDN going to manage both the overlay and the underlay network?
SDNCentral: The question here is whether SDN is needed to manage the underlay network. In a pure overlay model, the underlay network is managed separately and with normal management processes. SDN or network virtualization platforms manage the overlay separately. In the future, we expect to see SDN controllers tie together overlay and underlay management for improved resource utilization, better visibility and faster troubleshooting.
Nuage: A great question, to which today’s VPNs are a great corollary. They are, in full effect, overlay abstractions that interact quite closely with the underlying infrastructure to ensure that services can be assured and the right level of visibility is achieved. Central to addressing your question are several others that must be pondered, including: 1) Can overlay networking scale at data center capacities? 2) Could a correlation between overlay and underlay networks improve the services offered in a data center environment? And 3) Are there open standards that can achieve this, or is there a need for some new protocols or technologies? For a detailed response, please see a blog from our co-founder and CTO Dimitri Stiliadis here.
What’s the “wild card” for SDN in the next year or two? Meaning, what’s the part that’s likely to spring up and become really important?
SDNCentral: Historically, while most people understand the inherent importance of the SDN controller and the strategic significance, the maturity of the controller concept has some ways to go. I believe that the SDN controller models will go through some transition as service providers and enterprises attempt to deploy them with specific applications. This will happen both on the commercial and open-source fronts. Today’s frameworks as mostly single applications and single user and until we get to a multi-application framework that supports multi-tenancy, it will be hard to achieve the full benefits of SDN.
In addition, hybrid deployments, or co-existence with existing systems, will gain increasing importance as network operators attempt to roll out some type of SDN. SDN as the “exception” model within common frameworks like OpenFlow’s Normal mode or I2RS or proprietary APIs from vendors will gain more traction.
Nuage: One of the key areas that will become increasingly important is systems integration. As customers select best-of-breed elements to implement their version of an open cloud, some pieces will have been architected to work more readily with virtualized appliances and disparate hardware platforms, and in cross-hypervisor and cross-CMS environments, more so than others. As pieces of the puzzle evolve at different rates, the responsibility of keeping the elements functioning seamlessly can become a challenge. Highly vertically integrated systems from a single vendor reduce this risk, but can not be the answer if the promises of an open cloud and the benefits of innovation are to be realized. For that reason, the right platforms are key, and integration will certainly be one of the big wildcards in the year(s) ahead.
For Nuage, what part of what you’re saying do customers relate to the most? Is it the automation?
Nuage: Automation is certainly a major attribute that translates to ROI and delivers benefits in customer satisfaction. Some have said that “you can’t sell automation to those doing the manual labor,” but you can certainly sell them better tools. Otherwise, gardeners would categorically reject lawnmowers, which we know is not true. Indeed, they embrace better and more efficient tools that help them satisfy more customers more quickly.
The same is true with the benefits of policy-driven network auto-instantiation as opposed to manual device-by-device-by-device configuration. But besides policy-based automation and IT-friendly abstraction of network resources, what really resonates with our customers is the ability to work elegantly in their hybrid environments: cross-hypervisor, cross-CMS, agnostic to network hardware, to deliver the benefits of SDN across all of their data center assets, whether they are virtualized or not.
Can you virtualize service providers’ core routers?
Nuage: In theory, you can virtualize pretty much anything, but just because you can doesn’t mean you should. The reality is that there are functions that lend themselves strongly to virtualization, and those where you quickly hit diminishing returns.
The more compute-intensive a function is, the more readily the benefits of virtualization can be realized and reaped. Those are the functions (like IMS, for instance) that make the most sense to virtualize ASAP. On the other hand, the more bandwidth-intensive a function is, the more you need careful analysis to ensure the benefits are clear. They often don’t materialize, because you hit diminishing returns as you employ servers for I/O intensive functions. That’s certainly true today given the relative cost/performance curves of GPUs and throughput-optimized network processors.
For functions required in core routing, for instance, what is done today in one chipset on one line card would require several racks of servers to achieve. That is not a tradeoff that many are willing to even consider. We expect that trend to continue, but will look continuously at the evolution curves to dictate new functions that are ripe for virtualization.
Can you elaborate on the benefits and requirements for extending MPLS into the hypervisor/VMs?